Works with Figma. Copy layers there and back with a plain copy and paste. A token brings the images along.
Tasks
SPECSSpec tasks of the projects open in tabs. Click a task to see it on the canvas.
Gigma, an open-source design editor, works like Figma without the internet and lets your AI assistant draw on the same canvas.
Everything you need to design with AI:
Vector editing·Auto layout·Components·Spec nodes·Guidelines·MCP server·PDF import
Gigma is a vector design editor with auto layout, components, styles and a command palette. On top of that it has what a designer needs to work with an AI and with Figma.
Works with Figma. Copy layers there and back with a plain copy and paste. A token brings the images along.
An MCP server, built in. Gigma listens on your own machine and gives an assistant 62 tools: it reads layers and styles, takes screenshots and draws on the canvas.
Spec nodes. A task is a card on the canvas. Materials come in on its left, the frames to work in go out on its right, and the assistant follows the wires.
Guidelines. The colours, fonts, logos and rules of a brand, shared by every project and read by the AI before it draws.
Claude, Codex or any MCP client builds while you work, on the same canvas. Every change lands in the same history, so one Ctrl+Z takes back anything you do not like.
Wire a task to the frames it is about, and materials into it.
One brand for every project, read by the AI before it designs.
Every AI step is an ordinary edit in your history, never a black box.
Gigma runs an MCP server on your machine. The assistant reads layers, styles and screenshots, then draws back on the canvas. Use it from Claude, Cursor, VS Code or LM Studio.
Spec tasks of the projects open in tabs. Click a task to see it on the canvas.
Frames, text, shapes, styles, components, export. The whole inspector.
Screenshots and exports of any layer, so it checks its own work.
AI edits are normal history entries. Ctrl+Z works as always.
Tasks and rules wired to frames are handed to the AI where it works.
Exact colours, fonts and logos, with transparency, not a guess.
The server listens on 127.0.0.1. Your files never leave the machine.
Everything a designer reaches for, in one window. Every tool is also a command an AI can call.
PDF import. Pages become frames, text and shapes become layers.
Pen and curves. Draw paths, then bend every point.
Export. PNG, SVG and PDF, at the scale you need.
Storage. See what takes space on the disk and clear it.
⟨/⟩Code blocks. HTML and CSS rendered as a layer.
AI cursors. Every assistant gets its own cursor and name on the canvas, so you see who is working and where.
Share as a file. Projects, guidelines and plugins travel in formats of their own.
A plugin is a web page with a manifest, zipped into a .gplug file. It runs in a sandbox and talks to Gigma through the very commands the AI uses.
Plugins in Gigma can put a screen into a phone frame, give layers a neon glow or bring a snapshot of a whole website onto the canvas.
With one HTML file, a plugin can read the selection, create layers, use fetch with permission and keep its own settings, through the very commands an assistant calls over MCP.
<!-- index.html of a plugin --> <button id="go">Round the selection</button> <script> go.onclick = async () => { const nodes = await gigma.selection() await gigma.undoGroup(async () => { for (const n of nodes) await gigma.command('update_node', { id: n.id, cornerRadius: 16, }) }) } </script>
Write it by hand or give the plugin docs to an AI along with your idea. Drop the .gplug file into Settings, Plugins, and it is ready to run from Ctrl+K.
Free and open source. Your projects stay yours. The AI is optional and always under your control.